Self-Driving Cars: CV & ML
During my sophomore spring at Yale, I took a class called CPSC 335: Theory and Practice of Self Driving cars. The intent was to work with embedded systems in actual, miniature, self-driving cars, but when COVID-19 restricted access to our lab space, we had to go virtual, working with simulated cars on virtual tracks.
The projects for the course involved implementing two different steering methods for our self driving cars: one based on computer vision techniques and one based on machine learning.
First, I used computer vision techniques to track the lanes and keep the car centered within them, and used a PID controller to steer. To do so, we used the python implementation of CV2, using the hough transform as well as various filters to find the lanes. Secondly, I used behavioral cloning to predict an optimal steering angle. For this, I used a Convolutional Neural Network, built out in Keras.